Gröbner Bases. eliminating the leading term Buchberger s criterion and algorithm. construct wavelet filters

Transcription

1 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination MCS 563 Lecture 6 Analytic Symbolic Computation Jan Verschelde, 27 January 2014 Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

2 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

3 S-polynomials to eliminate the leading term The leading term of a polynomial f is denoted by LT(f). Any term is the product of a coefficient and a monomial, the leading monomial of is denoted by LM(f). LCM(x a, x a ) is the least common multiple of x a and x b. To eliminate the leading term of two nonzero polynomials p and q, we compute an S-polynomial (S = Subtraction): S(p, q) = LCM(LM(p), LM(q)) LT(p) p LCM(LM(p), LM(q)) LT(q) q. If p and q belong to the same ideal I, then S(p, q) I. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

4 an example Applying S(p, q) = LCM(LM(p), LM(q)) LT(p) p LCM(LM(p), LM(q)) LT(q) q to p = xy + 1 and q = y 2 1 using > lex leads to S(p, q) = LCM(xy, y 2 ) (xy + 1) LCM(xy, y 2 ) xy y 2 (y 2 1) = y(xy + 1) x(y 2 1), as LCM(xy, y 2 ) = xy 2 = x + y We used p and q to define an ideal I for which the result of the division algorithm depended on the order. With S(p, q) I we add leading terms to the basis for I. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

5 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

6 a Gröbner basis for an ideal A set of polynomials g is a Gröbner basis for an ideal I if 1 I = g and 2 the leading terms of g generate the ideal of leading terms of the polynomials in I, i.e.: LT(g) = LT(I). Theorem (Buchberger s criterion) A set g = {g 1, g 2,..., g s } is a Gröbner basis if and only if for all pairs g i and g j, i j, the remainder of the division of S(g i, g j ) by g equals zero. This criterion leads to an algorithm for a Gröbner basis. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

7 the Buchberger algorithm The Buchberger algorithm to compute a Gröbner basis: Input: f = {f 1, f 2,..., f N }, I = f. Output: g = {g 1, g 2,..., g s }, LT(I) = LT(g). g := f ; repeat h := g; for each pair (p, q), p q, p, q g do S := S(p, q); r := remainder of S after division by g; if r 0 then g := g {S}; end if; end for; until g = h. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

8 using Macaulay 2 $ M2 Macaulay2, version with packages: ConwayPolynomials, Elimination, IntegralClos PrimaryDecomposition, ReesAlgebra, SchurRing i1 : R = QQ[x,y,MonomialOrder => Lex]; i2 : I = ideal(x^2 + 1,x*y - 1); o2 : Ideal of R i3 : G = gens gb I o3 = y2+1 x+y We see that {y 2 + 1, x + y} is a Gröbner basis for the ideal x 2 + 1, xy 1 with the lexicographical order. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

9 some more Gröbner basics Buchberger s algorithm generalizes Euclid s algorithm for the GCD and row reduction for linear systems. With a Gröbner bases, the division algorithm solves the ideal membership problem. A Gröbner basis g is called reduced if 1 the leading coefficient of every polynomial in g is 1; and 2 for all p g, no monomial of p lies in LT(g \ {p}). Fixing a monomial order, any nonzero ideal has a unique reduced Gröbner basis. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

10 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

11 filtering a signal A filter with input {x n } and output {y n } is completely determined by {h n }, the impulse response. x n h n y n With the convolution operator, we compute y = h x = k x k h n k. Via the Z -transform, Z({x n }) = n x n z n, h x becomes Y(z) = H(z)X(z), with X(z) = Z({x n }), H(z) = Z({h n }), and Y(z) = Z({y n }). Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

12 filter design The function H(z) is called the transfer function of the filter. We design a filter by determination of H. Example of conditions on the transfer function: 1 h 2 = h 3, h 1 = h 4 ; 2 (z + 1) 2 divides H(z); 3 h n h n 2k = δ(k), n with δ(k) = 1 if k = 0, δ(k) = 0 if k 0. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

13 a polynomial system Appending the equation h 0 + h 1 + h 2 + h 3 + h 4 + h 5 + h 6 + h 7 1 = 0 h 2 h 0 + h 3 h 1 + h 4 h 2 + h 5 h 3 + h 6 h 4 + h 7 h 5 = 0 h 6 h 2 + h 4 h 0 + h 5 h 1 + h 7 h 3 = 0 h 6 h 0 + h 7 h 1 = 0 h 0 h 2 3h 4 5h 6 + 6h 7 + 4h 5 + 2h 3 = 0 h 1 + 3h 3 + 5h 5 + 7h 7 6h 6 4h 4 2h 2 = 0 h 2 h 3 = 0 h 1 h 4 = 0 h 0 + h 1 + 2h 2 + 3h 3 + 4h 4 + 5h 5 + 6h 6 + 7h 7 A = 0 leads to a more compact Gröbner basis. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

14 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

15 cancellation of leading terms Lemma 1 Let f 1, f 2,...,f N C[x] be such that LM(f i ) = x a, for all i = 1, 2,...,N. N Consider f = γ i f i, for γ i C. i=1 If LM(f) < x a, then f is a linear combination of the S-polynomials S(f i, f j ), 1 i j N. Proof. Denote LT(f i ) = c i x a for c i C \ {0}. For all i, j: LM(f i ) = LM(f j ): S(f i, f j ) = 1 c i f i 1 c j f j. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

16 telescoping sums proof continued As LM(f) < x a : N γ i c i = 0. i=1 Using telescoping sums in case N = 3: f = γ 1 f 1 + γ 2 f 2 + γ 3 f 3 = γ 1 c 1 ( 1 c 1 f 1 ) + γ 2 c 2 ( 1 c 2 f 2 ) + γ 3 c 3 ( 1 c 3 f 3 ) = γ 1 c 1 ( 1 c 1 f 1 1 c 2 f 2 ) + (γ 1 c 1 + γ 2 c 2 ) +(γ 1 c 1 + γ 2 c 2 + γ 3 c 3 )f 3 = γ 1 c 1 S(f 1, f 2 ) + (γ 1 c 1 + γ 2 c 2 )S(f 2, f 3 ), because γ 1 c 1 + γ 2 c 2 + γ 3 c 3 = 0. The extension for any N is clear. ( 1 f 2 1 ) f 3 c 2 c 3 Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

17 termination of division Lemma 2 For any p C[x] and f = (f 1, f 2,...,f N ), f i C[x] for i = 1, 2,...,N on input, the division algorithm terminates yielding q 1, q 2,..., q N, r C[x] : p = q 1 f 1 + q 2 f q N f N + r. ( ) N Moreover: LM(p) = max max (LM(q i)lm(f i )), LM(r). k=1 Proof. We first show that the division algorithm terminates. Observe that at each stage of the division algorithm we subtract from r (initialized with p), producing a sequence of polynomials r 0 = p, r 1, r 2,... Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

18 exploiting monomial orders To obtain r i+1 we subtract LT(r i) LT(f k ) from r i, so we have: LM(r i+1 ) < LM(r i ), for all i in the sequence. This sequence must terminate for a monomial order < where every set of monomials has a smallest element. ( ) N To show LM(p) = max max (LM(q i)lm(f i )), LM(r), first recall that k=1 as the algorithm terminates: LM(r) LM(p). To compute the q i s we collect terms LT(r) LT(f k ) LT(r). Therefore: LM(q i )LM(f i ) LM(p) and we are done. where LT(r) LT(f k ) g k cancels Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

19 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

20 proof of Buchberger s criterion Theorem (Buchberger s criterion) A set g = {g 1, g 2,..., g s } is a Gröbner basis if and only if for all pairs g i and g j, i j, the remainder of the division of S(g i, g j ) by g equals zero. Proof. The of the theorem follows from S(g i, g j ) g. For the direction, let f I. While we may write f in many ways, we choose this representation of f : f = s i=1 h i g i, h i C[x] for which x a = s max i=1 LM(h ig i ) is least. If LM(f) = x a, then LT(f) LT(g) and we are done. Otherwise,... Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

21 deriving a contradiction Otherwise we rewrite f into a representation f = s h i g i, h i C[x] where i=1 s max i=1 LM( h i g i ) = x a, contradicting our first choice (x a is least among all representations for f ), leaving only LM(f) = x a. To derive the contradiction, take H = { i g i g : LM(h i g i ) = x a } and consider p = i H LT(h i )g i. We have LM(LT(h i )g i ) = x a, for all i H and LM(p) < x a, because LM(f) < x a. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

22 applying Lemma 1 Application of Lemma 1 yields coefficients c ij C: p = c ij S(LT(h i )g i, LT(h j )g j ). i,j H,i j As LCM(LM(h i g i ), LM(h j g j )) = x a : S(LT(h i )g i, LT(h j )g j ) = = where x b = LCM(LT(g i )), LT(g j )). x a LT(h i g i ) h x a ig i LT(h j g j ) h jg j x a LT(g i ) g i xa LT(g j ) g j = xa x b S(g i, g j ). The expression for S(LT(h i )g i, LT(h j )g j ) shows: if S(g i, g j ) reduces to 0, then also S(LT(h i )g i, LT(h j )g j ) reduces to 0. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

23 applying Lemma 2 Collecting quotients from the division algorithm: s S(LT(h i )g i, LT(h j )g j ) = h ijk g k where by the second formula of Lemma 2: k=1 s max k=1 LM( h ijk LM(g k )) = LM(S(LT(h i )g i, LT(h j )g j )) < max(lm(h i g i )LM(h j g j )) = x a. Substituting the expressions back into g we get the representation for f which gives the contradiction. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

24 Gröbner Bases 1 S-polynomials eliminating the leading term Buchberger s criterion and algorithm 2 Wavelet Design construct wavelet filters 3 Proof of the Buchberger Criterion two lemmas proof of the Buchberger criterion termination and elimination Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

25 the Buchberger algorithm terminates Showing that this algorithm terminates also shows the Hilbert basis theorem, i.e.: any ideal has a finite basis. The key observation is that as long as the repeat loop does not terminate, we augment g with a nonzero polynomial S = S(p, q) for which LM(S) < LM(p) and LM(S) < LM(q), with respect to the term order <. Compared to h, we thus have that LT(h) LT(g). So as long as the loop runs, we create a chain of monomial ideals which cannot stretch for ever. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

26 Elimination Ideals Consider again a system of homogeneous linear equations. Applying row reduction to bring such a system into triangular form can be written in terms of taking S-polynomials. For an ideal I in C[x], x = (x 1, x 2,...,x n ), the kth elimination ideal is I k = I C[x k+1,...,x n ]. So I k consists of all polynomials in I for which the first k variables have been eliminated. Theorem (The Elimination Theorem) Let g be a Gröbner basis for an ideal I with respect to the pure lexicographical order x 1 > x 2 > > x n. Then the set g k = g C[x k+1,..., x n ] is a Gröbner basis of the kth elimination ideal I k. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

27 proof of the Elimination theorem Proof. To prove this theorem, we must show that LT(I k ) = LT(g k ). By construction, LT(g k ) LT(I k ), so what remains to show is that LT(I k ) LT(g k ). For any f I k, we must then show that LT(f) is divisible by LT(p) for some p g k. As f I: LT(f) is divisible by LT(p) for some p g. Since f I k, the only variables in f are x k+1,..., x n. Because of the lexicographic order: if LT(p) C[x k+1,...,x n ], then all other terms of p also C[x k+1,...,x n ]. Thus the p for which LT(p) divides LT(f) belongs to g k. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

28 Summary + Exercises We gave a definition for the Gröbner basis, explained Buchberger s criterion and algorithm. Exercises: 1 Solve the system for filter design. Use Maple or Sage to create a lexicographical Gröbner basis. Verify that by adding one more equation, the resulting Gröbner basis is more compact. How many real solutions do you find? 2 Apply Buchberger s algorithm by hand (you can use a Maple worksheet to compute all S-polynomials) to the ideal generated by the equations {x x 2 2 1, x 1x 2 1} using a pure lexicographical monomial order. 3 Show that for two systems f(x) = 0 and g(x) = 0: if f = g, then their solutions are the same. Give an example of a case for which the opposite direction does hold. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

29 more exercises 4 Consider the example f(x, y) = { x 2 + ǫxy + y 2 1 = 0 y 3 3x 2 y = 0 for ǫ 0. Although the solution set varies continuously with ǫ, we will verify that a Gröbner basis cannot be a continous function of ǫ. Use Maple or Sage for the following calculations: 1 Make a plot of the two curves defined by the polynomials in the system. Justify why all intersection points are well conditioned roots. 2 Compute Gröbner bases for various values of ǫ and examine the growth of the coefficients as ǫ gets smaller. 3 Compute a Gröbner basis where ǫ is a parameter. Interpret the results. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

30 and more exercises 5 The twisted cubic is a curve in 3-space defined by (x 1 = t, x 2 = t 2, x 3 = t 3 ), for a parameter t. Equivalently, the equations x1 2 x 2 = 0 and x1 3 x 3 = 0 defined the twisted cubic in implicit form. The surface of all lines tangent to points on the twisted cubic is x 1 = t + s x 2 = t 2 + 2ts (1) x 3 = t 3 + 3t 2 s, for parameters s and t. Compute a lexicographical Gröbner basis using a monomial order that eliminates s and t. Find an equation for the surface that defines all tangent lines to the twisted cubic. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31

31 one last exercise 6 With a lexicographical Gröbner basis and a solver for polynomials in one variable we can solve zero dimensional polynomial systems, systems that have only isolated solutions. Write a procedure in a computer algebra system that takes on input a lexicographical Gröbner basis and computes all solutions by applying the univariate solver repeatedly and substituting the solutions into the remaining equations. For a numerical solver, show that the working precision must be sufficiently high enough as the solver progresses, considering the example of exercise 4. Analytic Symbolic Computation (MCS 563) Gröbner Bases L-6 27 January / 31